> [in 2020/2021] the dominance of autoregression was not as well-established as it is today: GPT-3 had turned some heads, but the ‘ChatGPT moment’ wouldn’t come until late 2022
I disagree with this. Decoders were absolutely dominant in 2020 for chat. GPT2 was considered too dangerous to release, and I remember scrambling to get on the GPT3 waitlist. It worked.
(The only exception I will make is encoder-decoder models which now are often done by decoder-only.)
But what made it go mainstream was RL. RLHF at first, then other improvements like DPO that were less of a pain in the ass to set up. Adding diffusion on top of that would be an even bigger pain in the ass.
Before ChatGPT there really wasn’t much of a concept of pre-training and post-training. It was all pre-training. Post training was what made the bots conversational and not just “continuing the thing you wrote to them”.
So in short, diffusion never took off because it was just a more complicated way to generate tokens, and the real problem was getting tokens in the right distribution.
show comments
2001zhaozhao
I would love to see models that can think at different rates and also output a thinking scratchpad alongside output text instead of before all output.
Right now models need to rely on less legible compressed CoT to get high intelligence per token/step, but with diffusion they would just need to output more tokens per step instead.
p1esk
It’s refreshing to read something not AI generated.
vatsachak
I feel like there is still low hanging fruit on the auto regressive LLMs; the encoder
Marchant_hq
CDLMs sound promising for smoother, more coherent text generation. Excited to see how they tackle the token-level discontinuities.
ovin_dal
Diffusion models for language felt inevitable. Imagine the creative potential once these mature beyond current limits.
ainch
A great read - as with all of Sander's diffusion posts.
NickNaraghi
I wonder if we’ll get something like CDLMs for automated harness engineering, sort of piloting the LLM underneath.
show comments
amelius
"Attention is all you need" should be renamed into "Attention is sufficient but not necessary".
> [in 2020/2021] the dominance of autoregression was not as well-established as it is today: GPT-3 had turned some heads, but the ‘ChatGPT moment’ wouldn’t come until late 2022
I disagree with this. Decoders were absolutely dominant in 2020 for chat. GPT2 was considered too dangerous to release, and I remember scrambling to get on the GPT3 waitlist. It worked.
(The only exception I will make is encoder-decoder models which now are often done by decoder-only.)
But what made it go mainstream was RL. RLHF at first, then other improvements like DPO that were less of a pain in the ass to set up. Adding diffusion on top of that would be an even bigger pain in the ass.
Before ChatGPT there really wasn’t much of a concept of pre-training and post-training. It was all pre-training. Post training was what made the bots conversational and not just “continuing the thing you wrote to them”.
So in short, diffusion never took off because it was just a more complicated way to generate tokens, and the real problem was getting tokens in the right distribution.
I would love to see models that can think at different rates and also output a thinking scratchpad alongside output text instead of before all output.
Right now models need to rely on less legible compressed CoT to get high intelligence per token/step, but with diffusion they would just need to output more tokens per step instead.
It’s refreshing to read something not AI generated.
I feel like there is still low hanging fruit on the auto regressive LLMs; the encoder
CDLMs sound promising for smoother, more coherent text generation. Excited to see how they tackle the token-level discontinuities.
Diffusion models for language felt inevitable. Imagine the creative potential once these mature beyond current limits.
A great read - as with all of Sander's diffusion posts.
I wonder if we’ll get something like CDLMs for automated harness engineering, sort of piloting the LLM underneath.
"Attention is all you need" should be renamed into "Attention is sufficient but not necessary".